AIGC with integrating real-world physical information
With the introduction of prompt and reinforcement learning from human feedback(RL-HF),generative artificial intelligence can accelerate the experimental process and provide high-preci-sion prediction and has been widely used in physics,chemistry and other science and technology fields,which has gradually revolutionizing the traditional model of scientific research.However,starting from only a limited number of labeled samples,training a robust generative artificial intelli-gence(GAI)that can generalize to arbitrarily similar scenarios still seems out of reach.This challenge essentially stems from the fact that physical information has not been fully integrated in existing artifi-cial intelligence generative content(AIGC)practices.The cognitive ability of AIGC on the physical level has been hindered by objective factors such as insufficient computing resources in the past,and has not been able to make significant progress.Nowadays,with OpenAI,Google,and others demon-strating the smart advances that come with hardware upgrades,the above situation is expected to change.This paper comprehensively reviewed the latest research developments in the integration of physical information and AIGC,including multiple sources of physical information,integration frame-work and specific case studies.The frontier challenges and potential development opportunities in this field were summarized.
NETL
NSTL
万方数据
